[Kde-perl] RE: Up and mosly running on SunOS

Germain Garand germain at ebooksfrance.org
Fri May 9 17:35:35 CEST 2003


Le Vendredi 09 Mai 2003 14:49, jim cowan a écrit :
> >necessary? (I can't see how this is related to the
> >missing vtbl symbol you reported?)
>
> This looks like a feature of the Sun C++, the clue
> was in:
> http://docs.sun.com/source/816-2459/Ch1.Intro.html#pgfId-6720
>

mmh, I see... eventually, wether its wrong mangling or not, it looks like a 
template instanciation problem for Qt.xs l.49 & l.50

It seems it tries to link to an existing instance, instead of creating it with 
the template... or is Forte's template support limited to a few predefined 
types?

If you had the time, could you try to define Smoke::Index as a short again in 
libsmoke, and then
1) add an explicit  #include <qasciidict.h> at the top of Qt.xs 
or if it fails
2) change AsciiDict<Smoke::Index> to AsciiDict<short>?


> which sort of describes their name mangling. The
> Qt we got was 3.1.1, built on SunC++ by another in
> my group. nm'ing and greping around led me to believe
> that this was the problem.
>
> >I've already seen those magic killbackref panics.
> >It seems to be a perl 5.6.0 bug with deleted weak
> >references that are still pointed-to by their target,
> >and I see no easy way to fix those. I'll try to dig
> >a bit more though.
> >Did they ever happen to crash a program or is it
> >always in global destruction?
>
> Always on global destruction as nearly as I can tell.
>

OK, I can probably fix this one by manually undefing the offending reference 
to the weak reference... will try that.

Germain

> Thanks for the support. I'll get the diffs to you
> today or next week.
>
> thanks
> jim cowan



More information about the Kde-perl mailing list